PySpur-Dev/pyspur

Features

• AI Workflow Builders - Visual drag-and-drop platform for designing, testing, and deploying AI agent workflows with custom nodes and tools.
• Agent Pipeline Designers - Ships a visual drag-and-drop interface for designing, running, and iterating on agent workflows.
• Agentic Workflow Construction - Supports cyclic workflow graphs for iterative tool calling and agent reasoning loops.
• Visual Editors - Provides a visual drag-and-drop editor for designing and testing multi-step AI agent workflows.
• AI Model Integrations - Provides unified nodes connecting to multiple AI vendors and tools for rapid model comparison.
• Multi-Vendor Switching Interfaces - Ships unified nodes that let users switch between AI vendors to compare performance without reconfiguring.
• Visual Designers - Provides a visual playground for rapid prototyping of AI agent workflows.
• Document Grounding - Retrieves relevant information from user-uploaded documents and provides it as context to AI models.
• Retrieval-Augmented Generation - Passes queries through a retriever node to fetch relevant document chunks for grounded LLM responses.
• Conversational Session Management - Maintains per-user conversation history across multi-turn interactions by storing session data tied to workflow runs.
• Workflow API Endpoints - Exposes workflows as REST endpoints that accept input and return results for external integration.
• Human Approval - Pauses workflow execution at designated breakpoints for human review and approval.
• Ingestion Pipelines - Ingests files, extracts text, splits into chunks, and stores each chunk with source metadata for search.
• Multi-Model AI Orchestrators - Unified nodes that connect to multiple AI providers for rapid model switching and performance comparison.
• Retrieval Augmented Generation Pipelines - Creates RAG pipelines that parse, chunk, embed, and retrieve documents to ground AI responses in private data.
• Retrieval Augmented Generation Platforms - Ships an integrated platform for document ingestion, chunking, embedding, and vector search to ground AI responses.
• Vector Data Management - Parses, chunks, embeds, and upserts data into vector indices for retrieval-augmented generation.
• Branch Merging Nodes - Provides dedicated nodes for merging parallel workflow branches into a single output.
• Workflow Definitions - Provides an endpoint to fetch a complete workflow definition and metadata by ID.
• Workflow Creators - Ships a REST endpoint to create new AI agent workflows from definitions or templates.
• Workflow Management APIs - Provides RESTful endpoints to create, update, and execute AI agent workflows programmatically.
• Conversational Agents - Creates workflows that maintain conversation history and manage multi-turn interactions with structured inputs.
• Python Decorator Registrations - Ships a Python decorator system for registering custom workflow nodes with category and display name metadata.
• Workflow Batch Iterators - Run a workflow once for each row in a dataset, mapping columns to inputs and writing results to a file.
• Document Ingestion Pipelines - Ingests files through a multi-stage pipeline of parsing, chunking, embedding, and upserting into vector indices.
• Vector Search Indexes - Searches a vector index with a query and returns the most relevant text chunks ranked by relevance score.
• Semantic Search - Converts document chunks into vector embeddings and performs semantic search to find relevant passages.
• Vector Indexing - Builds a searchable vector index from a document collection using an embedding model.
• Workflow Modifiers - Provides a REST endpoint to modify an existing workflow's structure, name, and description.
• One-Click API Exposures - Publishes a completed agent workflow as a one-click API endpoint for integration into external applications.
• Asynchronous Run Launches - Launches workflows asynchronously, returning a run ID for later status polling.
• Synchronous Run Executions - Executes workflows synchronously, blocking until the final outputs are returned.
• Node Graph Engines - Executes workflows by traversing a directed graph of typed nodes, each wrapping a callable with input/output schemas.
• Python Tooling - Converts Python functions into reusable workflow nodes using a decorator with metadata.
• Human-in-the-Loop Workflows - Pauses workflow execution at designated nodes, storing state until a human approves or overrides via API.
• RESTful Workflow APIs - Provides RESTful endpoints to create, run, and manage AI workflows synchronously or asynchronously.
• Workflow Routing - Implements conditional routing logic that directs incoming data to matching processing branches.
• Decorator-Based Tool Registrations - Provides a decorator-based mechanism to convert any Python function into a reusable workflow tool.
• Agent Evaluation Tools - Runs agents against real-world datasets to measure and assess their effectiveness before deployment.
• Custom Dataset Evaluators - Loads user-provided datasets of problems and ground-truth answers to test workflows on domain-specific cases.
• Human-in-the-Loop Workflows - Pause-and-resume mechanism that lets human operators approve or override critical workflow steps before continuation.
• Programmatic Evaluation APIs - Starts an asynchronous evaluation job that runs a specified workflow against a configured evaluation in the background.
• Multimodal Input Processors - Accepts PDFs, videos, audio, images, and text as inputs so agents work with diverse data types.
• Multimodal Processing - Handles video, images, audio, text, and code within agent workflows for diverse input types.
• Human-in-the-Loop Approvals - Pauses workflow execution for human review and resumes after approval or override decisions.
• AI Operations - Exposes RESTful endpoints to trigger and control AI-related operations programmatically.
• Benchmark and Evaluation - Runs AI workflows against datasets and academic benchmarks to measure accuracy and compare versions.
• Evaluation Benchmarks - Runs pre-built benchmarks from academic datasets to measure an AI workflow's reasoning and knowledge.
• Tenant Workflow Listings - Provides a paginated endpoint returning all existing workflows.
• Decorator-Based Node Creators - Adds new node types to the visual editor using a simple Python decorator.
• LLM-Based Summarizations - Routes input through web scraping or PDF conversion, then summarizes extracted text using an LLM.
• Batch Processing - Processes batches of inputs by providing a dataset ID and mini-batch size for efficient bulk execution.
• Dataset Management - Provides API endpoints to upload, query, and manipulate datasets used in AI workflows.
• Workflow Deletions - Provides an endpoint to permanently remove workflows and their associated test files.
• Workflow Run Management - Provides paginated retrieval of all workflow runs ordered by start time.
• Instance Cancellation - Cancels a running or paused workflow by its run ID.
• Run Status Inspections - Provides programmatic interfaces for checking the current state of an asynchronous workflow run.
• Synchronous-Asynchronous Execution Toggles - Lets users choose synchronous or asynchronous execution for workflow API calls.
• Execution Pausing - Resumes paused workflows by approving, declining, or overriding with human inputs.
• Pause Action Resolutions - Resumes paused workflows by approving, declining, or overriding with human inputs.
• Node Type Registries - Provides an API to retrieve schemas for all node types available in the workflow editor.
• Partial Graph Executions - Executes workflows partially, starting from a specific node with precomputed upstream outputs.
• Workflow-Bound Session Managers - Provides endpoints to manage conversation sessions tied to specific users and workflows.
• Workflow Accuracy Evaluators - Runs a workflow against a dataset of known-correct problems and reports the percentage of correct answers.
• Benchmarking Systems - Provides a benchmarking system that runs workflows against academic or custom datasets to measure accuracy.
• Runtime Node Registries - Discovers and registers custom node types at runtime through Python decorators for the visual editor.
• Long-Running Operation Polling - Runs workflows in background tasks, returning a run ID for status polling via REST endpoints.
• Low Code Interfaces - Graph-based editor for designing and managing LLM workflows.

Star history